🔥我靠！终于有大公司把AI模型后训练的“魔法”拆成可复现配方了！
This is a AI post classified by AGK Intelligence as AI research (research), kept by the AI Radar because it carries real work, not commentary.
🔥我靠！终于有大公司把AI模型后训练的“魔法”拆成可复现配方了！ Amazon团队直接公开从GLM-4.5-Air-Base出发的完整8阶段串行后训练pipeline，数据来源、奖励设计、筛选方法、训练配置、阶段结果全部写清楚，且大量用公开数据、无新增人工标注、无内部蒸馏教师！ 他们发现： 1️⃣串行8阶段流水线（每一步继承上一步参数）： SFT → Reasoning RL → Coding RL → Instruction-Following RL → General Agent → Coding Agent → Search Agent → RLHF 从基础能力到高级Agent，从硬可验证奖励逐步过渡到软judge奖励。 2️⃣不造新数据也能大幅提升：大量复用已公开数据集，附录详细标注来源，证明高质量公开数据+合理筛选就能打。 3️⃣核心原则： ▫️多样高质量SFT先打牢能力地板 ▫️难度过滤让RL prompt保持在有效学习区间 ▫️奖励可靠性决定阶段顺序（先硬规则/执行验证，再软LLM judge） ▫️工程与基础设施本身就是配方的一部分 4️⃣效果炸裂：Rufus-Air在多个基准上大幅超越智谱官方后训练的GLM-4.5-Air，与同规模开源模型竞争激烈。 后训练不再是黑箱！这套recipe直接拉高了开源可复现的天花板。 论文链接：https://arxi
Posted by SuSu_酥酥👅 (26.9k followers) 8 h ago · 107 likes · 5.6k views · view the original post on X. Kept by the AI Radar as AI research. Tools mentioned: arXiv.org.
More AI work like this
- No need for me to manually scan for new model releases anymore, since @paperswithcodee… — @NielsRogge
- We’re introducing ToyBlocks, an open simulated tabletop for controlled experiments in… — @allen_ai
- HotBench - Expert-authored tasks in medicinal chemistry, microbiology and dermatology to… — @SynBio1
- I wonder if you realize how interesting this is. — @SciTechera
- BREAKING: the best of 8 frontier models scored 37.5/100 on Science's Last Exam. On true… — @guilleflorvs
- Stanford dropped a 2-hour course on building LLMs from scratch — @RoundtableSpace
- Meet DINOv3 ViT-H/16+: a powerful image feature extractor from the DINOv3 family. With… — @HuggingModels
- the self-reported uplift from Model 4 is insane — @scaling01
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 65.6k posts from 5k X accounts over the last 14 days, 10.4k tools, 19 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-06 16:05 UTC. Full method.